Date: Mon, 27 Apr 1998 09:58:51 +0300 (IDT)
From: Eli Zaretskii <eliz AT is DOT elta DOT co DOT il>
To: Kbwms <Kbwms AT aol DOT com>
cc: djgpp AT delorie DOT com
Subject: Re: Running paranoia in Single Mode
In-Reply-To: <b07cbd70.35435f90@aol.com>
Message-ID: <Pine.SUN.3.91.980427095803.22490A-100000@is>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
Precedence: bulk


On Sun, 26 Apr 1998, Kbwms wrote:

> It makes little sense to me to claim that float values
> are checked, when the keyword "Single" is defined, only to discover that
> such is not the case.

Your conclusions are wrong, IMHO.  When `Single' is defined,
paranoia.c *does* test single-precision float computations, since all
the variables are declared float instead of double.  So when a value
is stored in a variable, its low-order bits are lost.  And since
paranoia.c should be compiled without optimizations, the results
*will* be different in the single-precision version.

Please note that the main purpose of paranoia.c is to reveal the
characteristics of the underlying FP arithmetics and the FP code
generated by the compiler, not to test the library functions (that's
the purpose of elefunt).  AFAIK, library functions are only used in
paranoia.c to help determining these FP characteristics.

So in my view, defining `Single' goes a long way towards accomplishing
the goal for which paranoia.c was written, for single-precision
computations.